On 2011-05-31, at 14:01 , Philip Homburg wrote:

> I would say that having an interface with two IPv4 addresses is not really in
> the model.

Maybe it is rather uncommon, but it is allowed and also supported by all major 
operating systems; just wanted to point that out.

> But SHOULD is a bit tricky in that context. If you assume that statefull
> DHCPv6 is expensive (compare to SLAAC) and you want to support extremely
> minimal devices, then SHOULD may just be a bit too strong.

Is DHCPv6 more expensive than DHCPv4? Since even the most minimalistic device 
with IPv4 support I know of has DHCPv4 support; however, you  might be 
referring to devices even more minimalistic than what I have in mind. I just 
would like to point out, that I already wrote a DHCPv4 client myself (for a 
software project) - not fully featured, it only cares for an IP address, a 
subnet mask, DNS servers, DNS domain and the lease time, all other options are 
pretty much ignored - and I was surprised how little code I needed for that. It 
is pretty robust and so far worked well with every DHCP server we ever found. 
It even supports renewing the lease, though it does not fully implement the 
state-flow of the RFC (which seems over complicated IMHO). So a device that is 
not able to run such a simple client must be really, very, very minimalistic 
and my impression is that DHCPv6 is much simpler than DHCPv4, since it is not 
an extension to BOOTP (which makes DHCPv4 harder to implement than
  would otherwise be necessary).

> I'm not sure that a bit that distinguishes manual configuration from DHCP is
> a good thing. SLAAC with EUID already has its bit.

I'm not advocating to make DHCP and manual configuration distinguishable, but 
to make SLAAC with Priv Ext distinguishable of DHCP and manual addresses:

Manual (Static) Address: U = 0, A = 0
DHCP Address: U = 0, A = 0
SLAAC: U = 1, A = 1
SLAAC + Priv Ext: U = 0, A = 1

> So the only thing that remains is how much faith you have in a random number
> generator. I think it is no problem, but we have to wait for large scale
> deployment to see if that is justified.

As a programmer, I don't believe in randomness but in predictability. You don't 
write code that  will only crash in one out of hundred billion cases if you can 
write code that will never crash for sure - even if the first case is so 
unlikely, that you may never see the code crash in your whole life.

However, we are losing the point here: With IPv6 I just fail to see the 
argument SLAAC vs DHCPv6, since there is no real reason IMHO for either/or and 
I fail to see how this is helpful if there are networks based on SLAAC and 
networks based on DHCPv6, this only adds unnecessary inconsistency. And people 
make it even worse by requesting a flag to control whether nodes may use 
privacy extension or not. Why not make it consistent: Every network is always 
SLAAC, even if there is a DHCP server, it is SLAAC. Privacy Extension is always 
possible to protect people's privacy if they feel like it. And only DHCP is 
optional, but when it is available, it does not replace SLAAC, it "completes" 
it and either hands out an IP address or not, but even when it hands out an IP 
address, it does not replace SLAAC. And thus I don't really see any need for an 
M or O flag in a RA message. Clients supporting DHCP will always try DHCP when 
a link comes up. Really, this traffic is so little compared to
  current network speeds, it's absolutely insignificant.

Regards,
Markus
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to